A fast algorithm for mining association rules in image

Association rules have been used in data mining applications to capture relationships present among attributes in large data sets. It can be adapted to capture frequently occurring local structures in images. The frequency of occurrence of these structures can be used to characterize texture. In order to mine the frequency patterns of texture, each image can be considered as one transaction. If image data mining drills down to pixel level, each pixel or its neighborhood can be taken as transaction too, and data mining was processed in all the transactions. In textural image, the frequent patterns are texture cells in fact. Because of different size of texture cells, multi-levels and multi-resolution data mining can be accomplished. One texture image has many texture cells so that the texture combined association rule can represent the texture feature; and segmentation can be accomplished based on texture combined association rules, image. The experimental results demonstrated that the combined association rules can represent both regular and random texture perfectly. Many experiments testify that 3 degrading ranks and 3×3 mask or 4 degrading ranks and 2×2 mask can mine the combined association rules which can represent the image texture perfectly. Simulation results using images consisting of man made and natural textures showed that combined association rule features performed well compared to other widely used texture features.